Marrying for Money: Evidence from the First Wave of Married

Women’s Property Laws in the U.S..∗

Peter KoudijsStanford GSB

Laura SalisburyYork University

September 2018

Abstract

Marriage can substitute for formal business contracts, especially in environments that lacka well established system of contract or corporate law. In such settings, marriage can facilitatethe efficient organization of labor and capital. In this paper, we explore the pooling of capital asan explicit motive for marriage. We measure the impact of a class of married women’s propertyacts introduced in the American South during the 1840s on assortative matching in the marriagemarket. These laws did not grant married women autonomy over their separate estate; theymerely shielded their property from seizure by their husbands’ creditors. This had the dualeffect of mitigating downside risk while restricting a husband’s ability to borrow against hiswife’s property; it also preserved the bulk of the wife’s assets as inheritance for the couple’schildren. Using a newly compiled database of linked marriage and census records, we showthat these laws were associated with an overall increase in assortative mating, suggesting thatthe ability to pool capital importantly contributed to the gains from marriage. At the sametime, there is considerable heterogeneity in the effect in different regions of the joint men’s andwomen’s wealth distribution. We provide an interpretation for these results.

∗Preliminary and incomplete.

1

1 Introduction

Marriage is, in part, an economic decision. By marrying, a couple can achieve economic outcomes

that are difficult to attain relying on market transactions alone. Analogous to the Theory of the

Firm, a marital contract can improve efficiency through the enforcement of implicit contracts and

by discouraging opportunistic behavior. The existing literature, starting with the seminal work

of Becker (1981), has emphasized the improvements in household production that are achieved

through marriage, focusing on married couples’ ability to exploit increasing returns through division

of labor.1 In this paper, we take a different perspective and consider the role of marriage in the

pooling of financial resources for business purposes. In particular, through marriage, couples can

pool assets to increase the amount of pledgeable collateral, improving access to credit. At the same

time, by strategically allocating ownership of marital property between spouses, households can

acquire additional protection from outside creditors, effectively substituting for limited liability.

Even though marriage can serve as a safeguard against opportunistic behavior, there is still scope

for moral hazard, as one partner can misappropriate funds. This is especially problematic if divorce

is a costly option or if property rights upon separation are not well defined.

The idea that marriage has traditionally served as a substitute for more formal business con-

tracts is not new, especially in historical contexts and developing countries that lack a rich system

of contract or corporate law. However, most of the existing literature has emphasized how marriage

can facilitate the efficient organization of labor. The role of capital is not well studied, and the main

purpose of this paper is to fill this gap in the literature.2 It is not straightforward to empirically

disentangle the different economic roles of marriage and isolate the role of capital. In this paper,

we take the following approach. We locate an institutional change that affects how capital can be

used within marriage and that affects the gains of marriage in a clear and predictable way. We

then measure its effect on partner choice. If the pooling of capital is indeed important, we expect

1See Weiss (1997) for an overview.2Historian Stephanie Coontz (2005) documents the evolution of marriage from a primarily economic to a primarily

emotional institution, noting that, “for much of history, marriage was not primarily about the individual needs anddesires of a man and woman and the children they produced. Marriage had as much to do with getting good in-lawsand increasing one’s family labor force as it did with finding a lifetime companion and raising a beloved child” (p.6). Fafchamps and Quisumbing (2005a, 2005b) and Rosenzweig and Wolpin (1985), among others, look at developingcountries. There is also a literature on the role of family networks in accessing credit; for a recent example, see Leeand Persson (2016).

2

assortative mating on familial wealth to change in a particular way.

We are especially interested in understanding how a couple’s ability to access the credit market

affects the gains from marriage. To this end, we exploit a unique institutional development in the

American South during the 1840s – the introduction of a specific class of married women’s property

laws – that affected the allocation of marital property and married couples’ interactions with credit

markets. Prior to the introduction of these laws, a woman’s property became her husband’s property

upon marriage. These laws altered this default, but in a very limited way. They did not give a

married woman the right to determine how her property was used, but, instead, shielded her assets

from seizure by her husband’s creditors. In addition, a husband could not dispose of his wife’s

assets, unless his own income and wealth were insufficient to provide for the family. As such, the

married women’s property laws both functioned as a form of bankruptcy protection and a vehicle to

reduce moral hazard on the part of the husband. Importantly, if a couple’s ability to access credit

is an important source of the gains from marriage, then these laws will have a clearly predictable

effect on the the relative value of certain types of marriages, which is heterogeneous in different

areas of the joint men’s and women’s wealth distribution (details follow).3

We study changes in marriage choice by looking at the tendency for people from similar so-

cioeconomic backgrounds to marry each other. The effect of married women’s property laws on

assortative mating depends on whether or not spousal capital becomes more or less “complemen-

tary” after the passage of a law. If spousal wealth becomes more complementary, then the relative

gains from rich men marrying rich women will increase, and assortative mating should become more

prevalent (Becker 1981). We model the way property laws affected a couple’s interaction with the

credit market, and how this in turn affected the relative complementarity of spousal assets. In the

absence of a property law, husbands’ and wives’ capital are perfect substitutes. If a wife’s property

3We also note that, because these laws predated modern divorce laws and did not allocate economic power towomen, they altered the way in which marital property could be pooled without having a first order effect on therelative bargaining power of spouses. There is a literature on the way in which bargaining power within the householdaffects spouses’ household labor input. For instance, Chiappori et al (2002) explore the impact of divorce laws on thedivision of labor between spouses, citing changes in bargaining power as the driving mechanism. Geddes and Lueck(2002) discuss the adoption of U.S. state statutes allowing women to own and control property. They argue that theadoption of these laws can be explained in part by increasing returns to women’s work: if women invest more effortin production when they hold property rights within the family, this may explain why male dominated legislatureswere willing to pass such legislation. Doepke and Tertilt (2009) propose a model in which assigning property rightsand increased bargaining power to married women increases investment in children’s human capital. While we expectthe effect on bargaining power to be minimal, we emphasize that a model in which property laws affect bargainingpower is unlikely to generate the type of heterogenous effect on the marriage market that we document here.

3

is protected, spousal assets become more complementary: only the husband’s capital can be used

to access credit, but only the wife’s capital can be consumed in the event of default. On the other

hand, by affecting borrowing behavior, the property laws also affect the level of consumption a cou-

ple can enjoy. If utility becomes less concave in consumption as consumption increases (as in most

conventional utility functions for risk averse agents), then husband’s and wife’s wealth becomes less

“substitutable” if the consumption a couple is able to achieve increases. In related work, Koudijs

and Salisbury (2016) show that the increased protection against creditors only benefitted couples

with relatively rich husbands who had sufficient collateral available to begin with; these couples

were able to increase borrowing, which led to increased consumption. Couples with relatively rich

women had a serious lack of pledgeable assets after the legal change and became credit constrained,

leading to decreased consumption. This implies heterogeneity in the impact on the marriage mar-

ket, with assortative mating becoming more common among couples with richer husbands and

poorer wives, and less common among couples with richer wives and poorer husbands. We show

that, while several alternative models deliver an increase in assortative mating overall, they do not

predict this type of heterogeneity.

This result illustrates how useful the property laws we analyze are for establishing the link

between a couple’s interaction with the credit market and the gains from marriage. Unlike married

women’s property laws passed elsewhere in the U.S. and later in the 19th century, these laws did

not allow married women to borrow money, so the portion of marital assets that are excluded

from interaction with the credit market is clearly identifiable. We argue that the effect of a law

on assortative mating depends on the ratio of protected to unprotected assets. Thus, if we know

that women’s assets are protected and men’s assets are unprotected, this gives us a clear prediction

about what heterogeneity in the marriage market should look like. We get no such clarity from

laws allowing married women to borrow independently of their husbands.

Using a new database of marriage records in the U.S. South from 1840 to 1851, we document

striking changes in marriage patterns after the passing of the laws that are consistent with this

simple framework. Using the 1840 census, we determine each partner’s socio-economic background.

We define “marriage markets” to be state-years, and we use a statistic based on Choo and Siow

(2006) to measure the systematic gains from assortative versus non-assortative matches. We inves-

tigate how this changed after the introduction of a married women’s property act. Because different

4

states passed laws at different times, we can include both state and year of marriage fixed effects

in our regressions. We show that married women’s property laws were associated with an overall

increase in assortative mating on wealth. However, this masks heterogeneity in different parts of the

joint distribution of men’s and women’s wealth. In particular, we show that laws tended to increase

the gains from assortative matching among couples with relatively richer husbands; however, they

tended to decrease the gains from assortative matching among couples with relatively richer wives.

This is entirely consistent with the theoretical effect of these laws, described above.

These results indicate that, in the absence of a modern system of contract and business law,

the institution of marriage has an important economic role in alleviating problems associated with

credit constraints and limited debtor protection. Our findings indicate that changing laws governing

marital property can materially affect marriage choices. This has interesting implications about

the development of marriage patterns and assortative mating over time. As the U.S. economy has

developed, and access to outside finance and limited liability has improved, some of the economic

motives for marriage have disappeared. We hypothesize that this has fundamentally changed the

way in which people have formed partnerships over time.

Our paper relates more generally to the literature on assortative matching in marriage markets.

There is evidence of a recent decline in marriage rates accompanied by an increase in assortative

matching on economic status (Choo and Siow 2006; Greenwood et al 2014). This has sparked new

interest in understanding the way economic institutions interact with marriage markets. Assortative

matching can be explained by different economic mechanisms. First of all, they can be the result

of random matching with search frictions. These models may generate assortative matching if

people with similar characteristics are more likely to encounter one another in the marriage market

(Adachi 2003). Second, non-random matching models posit that people have preferences for certain

traits in the marriage market. Assortative matching will occur in a frictionless setting with stable

matches if certain traits are universally preferred by both men and women – in this case, highly

ranked men will pair with highly ranked women, and lower ranked men will pair with lower ranked

women.4 Alternatively, if people prefer mates with similar characteristics to themselves, assortative

matching will also tend to occur when matches are stable.5

4See Chen et al (2013) and Olivetti et al (2015) for examples of such models.5Gale and Shapely (1962); Weiss (1997).

5

The fact that different marriage matching models have similar predictions for assortative match-

ing make it difficult to differentiate between these models. Hirtsch et al. (2010) show that assor-

tative matching emerges in online dating – a relatively frictionless setting – and argue that this

indicates that people have explicit preferences for similar mates in the dating market. Our paper

takes a different approach: we show that changes in marital property regimes generate changes in

assortative matching on economic status. Since these property regimes had no effect on marriage

matching institutions, this only makes sense if spousal economic assets enter directly into a person’s

utility function.

Finally, this paper adds to the literature on married women’s property laws in the United

States. This is a topic that has received much attention from economists and economic historians;

however, due to data limitations, it has been difficult to introduce pre-marriage characteristics

into any empirical analysis of these laws. In particular, it is difficult to observe pre- and post-

marriage socioeconomic characteristics of both halves of a couple, and to know whether a couple

was married before or after the passage of a married women’s property law. Most examinations

of the consequences of these laws have focused on their effect on women’s economic activity or

wealth holding, typically looking at state-level changes in these outcomes following the passage of

a property law. Kahn (1996) explores the effect of married women’s property laws on women’s

patenting, examining changes in the rate of patenting among women at the state level. Inwood

and Van Sligtenhorst (2004) look at changes in women’s property holding that occurred after the

passage of a married women’s property law in Ontario, Canada. Geddes et al (2012) analyze the

effect of property laws on children’s school attendance at the state level. Hamilton (1999) analyzes

choices of property regimes by married couples in 19th century Quebec, who could opt for separate

or community of property through prenuptial contracts. Koudijs and Salisbury (2016) analyze the

impact these property laws had on family investment decisions.

The rest of this paper is structured as follows. Section 2 provides more historical background.

Section 3 develops a simple model to describe the impact of married women’s property laws on

assortative mating. Section 4 discusses the newly constructed dataset that we use in this paper, and

section 5 describes the empirical strategy. Section 6 presents empirical results on the assortativeness

of matching. Section 7 discusses alternative mechanisms that may drive our results, and section 5

concludes.

6

2 Historical Background

Prior to the introduction of married women’s property acts, married women’s property was governed

by American common law, which dictated that virtually all property owned by a woman before

marriage or acquired after marriage belonged to her husband. The exception was real estate.

Although the fruits derived from real estate belonged to the husband (who could use this revenue

as collateral for a loan), the property itself was inalienable and was held in trust by the husband

for his wife. It was supposed to pass on to their children or otherwise would revert back to the

wife’s family (Warbasse 1987, p.9). In most of the states we consider in our empirical analysis

prenuptial agreements were problematic to enforce and therefore rare (Salmon 1986, p. xv). The

key difficulty lay in the dual legal system in the U.S. at the time. The dominant legal framework

was American common law. Under this system prenuptial agreements were not valid. To ’fix’ some

of the inequities of common law, a separate body of equity law had evolved. This branch of the law

did support prenups, but it was less well established and was administered in separate chancery

courts. This created two problems. First, as many southern states did not structurally report

equity cases, chancery judges often knew little of the equity jurisprudence. Second, there were few

courts that solely administered equity law. Usually, a judge mixed equity and common law cases.

As a result, decisions were rife with inconsistencies (Warbasse 1987, p. 165-6).

Warbasse (1987) suggests that the problems associated with equity law and prenuptial agree-

ments spurred the passing of state statutes modifying the common law to better protect women’s

assets within a marriage. These laws were introduced at different times indifferent states.6 The acts

can be broadly separated into four categories: debt relief, or acts that shielded women’s property

from seizure by husbands’ creditors but did not allow women to control their separate property;

property laws, or laws that allowed women to independently own and dispose of real and personal

property; earnings laws, which allowed women to control their own labour earnings; and sole trader

laws, which allowed women to engage in contracts and business without their husbands’ consent.

We focus on the first class of married women’s property acts (“debt relief”), which were enacted

in most southern states during the 1840s. Interestingly, the states that did not pass these law

6Information on married women’s property acts is compiled from a number of sources, including Kahn (1996),Geddes and Lueck (2002), Warbasse (1987), Kelly (1882), Wells (1878), Chused (1983) and Salmon (1982).

7

changes had the most well developed equity law systems, such as Virginia and Georgia (Warbasse

1987, p. 167). The timing of the passing of these laws coincided with a major recession, following

the Panic of 1837, which precipitated a large decline in cotton prices. This depressed land and slave

prices in the southern states, where the economy and financial system was based largely around

plantation agriculture (McGrane 1924). Historians argue that these laws were passed in response

to the economic hardship created by this recession, and the observation that men’s losses were also

being borne by their wives (Kahn 1996). At the time all loans were full recourse. If a husband’s

assets were not sufficient to cover a mortgage, for example, creditors could lay claim on all other

possesions a couple might have had, including a wife’s assets. For example, an article in the 1843

Tennessee Observer states that “the reverses of the last few years have shown so much devastation

of married women’s property by the misfortunes of their husbands, that some new modification

of the law seems the dictate of justice as well as prudence.” The Georgia Journal argued in the

same year that there is no good reason “why property bequeathed to a daughter should go to pay

debts of which she knew nothing, had no agency in creating, and the payment of which, with her

means, would reduce her and her children to beggary. This has been done in hundreds of instances,

and should no longer be tolerated by the laws of the land” (quoted in Warbasse 1987, p. 176-

177). This seems to have been a widespread sentiment, and even states that did not succeed in

passing a married women’s property act during the 1840s proposed them to the state legislation.

For example, Georgia failed to pass an act in 1843 by a margin of 18 out of 173 votes. Tennessee

did not pass an act until 1850, even though the issue had clearly been raised prior to this.

The first such law was passed in Mississippi in 1839, which merely sheltered a woman’s slaves

from seizure by her husband’s creditors; an additional law was passed there in 1846, securing

the income earned from her real and personal property to her separate estate. Alabama, Florida,

Kentucky, North Carolina, and Tennessee all passed similar property laws during the 1840s. Virginia

and Georgia did not pass laws during the period, and Louisiana and Texas were community property

states which kept property owned before marriage separate prior to the 1840s. Arkansas passed

a weak version of a property law in 1846, which was generally considered nothing more than a

strengthening of the equity tradition, which governs premarital contracts (Warbasse 1987). Table 1

contains a list of important legislative dates for each state that we use in our analysis. In all cases,

the statutes did not grant women the right to control their separate property; it was kept in a trust

8

administered by their husbands. As Kahn (1996) writes, “control remained with the husband, and

courts interpreted the legislation narrowly to ensure that ownership did not signify independence

from the family” (p. 361).

While the married women’s property acts passed in the South during the 1840s did not grant

women economic independence, they did place real constraints on the way in which this property

was used. As said, wifes’ assets were protected from from husbands’ creditors. At the same time,

a wife could not contract debt in her own name. Under common law a married women ( or ’feme

covert’) was legally unable to sign contracts; common law assumed that a family was a single legal

entity, led by the husband. The early married women’s property acts did not (yet) change this

feature of American common law. This put a wife’s assets in a special position: neither husband

nor wife could use them as collateral to obtain credit. In some states an exception was made to

furnish the household with ’common law necessaries,’ which included food and shelter. In general,

husbands and wives were allowed to jointly sell wife’s assets. However, this did not mean that

the ownership changed or that proceeds could be consumed. The proceeds from the sale had to

be reinvested as part of the wife’s separate estate. For example, an Alabama decision from 1857

maintains that, even if a wife’s property can be sold by a husband and wife jointly, the proceeds

“are to be reinvested in ’the purchase of other property’ not sold for money” (31 Ala. 39). The

statute was interpreted to protect a wife’s property “not only against third persons, but against

the husband himself.” This principle seems to have been broadly upheld in court.

A secondary motive for passing the married women’s property acts was the legislatures’ con-

cerns with the “character” of certain men. In 1846 the Alabama legislature commented that the

passing of a law would not only protect a women against a husband’s insolvency, but also against

his “intemperance or improvidence.” In 1839, a newspaper from Vicksburg, Mississippi argued,

somewhat less eloquently, that “the property of ladies should be guarded against the squandering

habits of a drunken and gambling husband. The ladies are virtuous and prudent creatures – they

never gamble, they never drink, and there is no good reason why the strong arm of legislation

should not be extended to the protection of the property they bring into the marriage bargain”

(quoted in Warbasse 1987, p. 150 and 170).

Of course, the extent to which these laws had any meaningful impact depends on the degree to

which women held property during this period. As women’s labor force participation was very low,

9

women’s property would have to come from family. The historical evidence suggests that women

frequently received real estate and personal wealth from their family. The first channel was dowry.

Though there is a serious lack in research on dowry in the Antebellum South, historical anecdotes

suggest that dowry was a frequent phenomenon. Thomas Jefferson’s wife, for example, received a

dowry of 132 slaves and many thousands of acres of land (Gikandi 2011). Auslander (2011) gives

numerous examples from Antebellum Greenwood county, Georgia of the transfer of slave property

in the form of dowry. The second channel was inheritance. After the American Revolution the

United States had done away with the British standard of primogeniture. In 1792 most US states

(including the South) had passed so-called intestacy laws that guaranteed that in the absence of a

will, sons and daughters would receive equal shares in the inheritance from their parents (Salmon

et al. 1987, p. 64-65; 83). There is very little evidence on the exact shares stipulated in actual

wills, but anecdotal evidence suggests that women could receive sizable inheritances, often in the

form of slaves (Warbasse 1987, p. 143-144; Brown 2006).7

3 Theory

We model the way in which the legal status of married women’s property affects the utility both

men and women derive from marriage. We focus on the impact on the relative value of marriages of

different types, which has implications about the degree of assortative mating on economic status.

We are especially interested in how married women’s property laws affected a couple’s interaction

with the credit market. By sheltering a wife’s assets from seizure by her husband’s creditors, the

law offers a household downside protection if the husband defaults on his debts; however, these

laws also prohibit a couple from borrowing against a wife’s assets, which limits a household’s access

to credit.

Intuitively, the impact on assortative mating hinges on whether these laws make husband’s

and wife’s wealth more or less substitutable. We proceed by showing that, by changing the way

a couple is able to access credit, the laws have a heterogeneous effect on assortative mating in

different regions of the joint men’s and women’s wealth distribution. When men are wealthy

relative to women, the laws tend to make men’s and women’s wealth more complementary through

7The tendency to will real estate to men seems to have been a national phenomenon in the first half of the 19thc.: see Salmon et al. (1987, p. 111) on the case of Bucks county in Pennsylvania.

10

this channel; when women are wealthy relative to men, the laws tend to make men’s and women’s

wealth less complementary.

3.1 Utility from Marriage

Men and women with wealth wM and wF , respectively, experience gains from marriage which can

be divided into three components: (i) a systematic component that derives from the consumption

of goods and services within marriage (UM (wM , wF ) + UF (wM , wF )); (ii) a systematic component

that does not derive from consumption but is correlated with assets (ψ(wM , wF )); and (ii) an

idiosyncratic component (φ). Thus, the value of a marriage between a man with wealth wM and a

woman with wealth wF who have idiosyncratic gains from marriage equal to φ is:

V (wM , wF , φ) = UM (wM , wF ) + UF (wM , wF ) + ψ(wM , wF ) + φ

Choo and Siow (2006) show that, if idiosyncratic match quality obeys an extreme value distribution,

the total systematic value from marriages between men of type i and women of type j is directly

proportional to the number of i, j marriages relative to the number of single men of type i and single

women of type j. Thus, the prevalence of certain marriages depends directly on their systematic

value.

In what follows, we model systematic utility U(wM , wF ) ≡ UM (wM , wF ) +UF (wM , wF ), which

accrues from the couple’s ability to invest and then consume premarital assets.

3.1.1 Statistic for Measuring Effect on Matching

Consider a marriage market with two types of men (wHM , wLM ) and two types of women (wHF , w

LF ).

Assortative mating in the marriage market will become more valuable – and thus more prevalent –

if the following expression increases:

∆U ≡ UHH + ULL − UHL − ULH

Here, UHH ≡ U(wHM , wHF ), and so on. Define Uij (i, j ∈ L,H) to be systematic marital utility

before the passage of a property law, and Uij to be systematic marital utility after the passage of

11

a property law.

Proposition 1 Suppose the following inequality holds for all x ∈ [wLM , wHM ] and y ∈ [wLF , w

HF ]:

∂2U(x, y)

∂x∂y>∂2U(x, y)

∂x∂y

Then, the following also holds:

UHH + ULL − ULH − ULL > UHH + ULL − ULH − UHL

Conversely, if the following holds for all x ∈ [wLM , wHM ] and y ∈ [wLF , w

HF ]:

∂2U(x, y)

∂x∂y<∂2U(x, y)

∂x∂y

Then, the following also holds:

UHH + ULL − ULH − ULL < UHH + ULL − ULH − UHL

Proof. Follows Becker (1981). See appendix for details.

In words, this means that the gains from assortative mating will increase when ∂2U/∂wM∂wF

increases, and these gains will decrease when ∂2U/∂wM∂wF decreases. So, in what remains, we

will characterize what happens to ∂2U/∂wM∂wF after the passage of a law.

3.1.2 A Model of Credit and Matching

Married women’s property laws affect the marriage market by altering a married couple’s interaction

with the credit market. Here, we model the effect of property laws on matching through this channel.

3.1.3 Setup

When a couple is married, the husband and wife invest their assets in a risky project with a

positive expected return. Project returns are then realized, and the couple consumes the return.

With probability 1/2, the project succeeds and yields a return of R > 1; otherwise, the project fails

and yields a return of R < 1. We define r ≡ R+R2 > 1, and ∆r ≡ R − R. Households can borrow

12

to scale up investment in the project. A portion of the return equal to RI can always be seized by

the creditor, where I is the total amount invested; however, the household can abscond with the

remainder. If the project has failed, there is nothing for the household to abscond with. But if the

project has succeeded, the household can abscond with ∆rI. We assume that there is a penalty

associated with absconding, so households who abscond with ∆rI can only consume β∆rI, where

β < 1. Thus, in order for a loan contract to be incentive compatible, the household must be better

off repaying its loan if the project succeeds than it would be if it absconded with the surplus. We

impose the following restrictions on parameters:

2(r − 1)

∆r< β < 1

RR− r > 0

The first restriction ensures that the incentive compatibility constraint is binding; the second

ensures that returns are not so risky that the household forgoes investment in the risky project and

instead chooses to hold assets in cash. We assume that both the husband and the wife consume

the proceeds of the household’s investment.8

3.1.4 No Separate Property for Women

Husbands gain full ownership of wF upon marriage, and they choose a loan size l to maximize their

own utility:

maxlU(l) =

1

2log cG +

1

2log cB

Here, cG is consumption of the married couple if the project succeeds, and cB is consumption of

the couple if the project fails. Assuming that l is risk free – which will certainly be the case, as a

risky loan would leave the couple with zero consumption if the project fails – these terms have the

following definitions:

cG = R(wM + wF + l)− l

cB = R(wM + wF + l)− l8Our key results are invariant to dividing marital output between spouses as opposed to allowing them both to

consume all of it.

13

Proposition 2 A household consisting of a husband with wealth wM and a wife with wealth wF

will choose the following loan size:

l∗ =RR− r

(R− 1)(1−R)(wM + wF )

Proof. See appendix.

Lemma 3 Given the household’s optimal choice of loan size, consumption in each state will be

equal to:

cG =∆r

2(1−R)(wM + wF )

cB =∆r

2(R− 1)(wM + wF )

Proof. See appendix.

3.1.5 Separate Property for Women

After a married women’s property law is enacted, the household can no longer use wF as collateral,

and wF cannot be seized in repayment of the households debts. Now, because wF cannot be seized

by creditors, the household may choose to contract a risky loan, such that (1 − R)l > RwM . If

it does this, it will pay the creditor R(wM + l) if the project fails, and ρl if the project succeeds,

where ρ > 1. This loan contract will need to satisfy the incentive compatibility constraint:

R(wM + l)− ρl ≥ β∆r(wM + l)

In addition, the lender requires a return equal to the risk free rate, which we set equal to 1 for

simplicity:

R(wM + l) + ρl = 2l

Whether or not the husband chooses to contract a risky loan will depend on the size of wF .

Namely, if wF is sufficiently small that the husband would prefer to consume more than RwF if

the project fails, he will opt for a risk-free loan, solving the same problem, and thus borrowing the

14

same amount, as before the passage of a property law.9 If wF is sufficiently large, the husband will

choose to contract a risky loan: he will solve the above maximization problem, but consumption in

each state is defined differently:

cG = R(wM + wF + l)− ρl

cB = RwF

Proposition 4 There exists a cutoff, Ω0, such that we obtain the following solutions for consump-

tion in each state:

1. wM/wF ≤ Ω0.

cG =R

λ(wM + λwF )

cB = RwF

where λ ≡ R(2−2r+β∆r)2β∆r < 1/2.

2. wM/wF > Ω0.

cG =∆r

2(1−R)(wM + wF )

cB =∆r

2(R− 1)(wM + wF )

Proof. See appendix.

3.1.6 Effect on Assortative Mating

We are interested in what happens to total systematic utility after the passage of a property law.

Recall that total utility is:

V (wM , wF , φ) = log cG + log cB + ψ(wM , wF ) + φ

9This must be feasible, since consuming more than RwF in the bad state implies borrowing against less than wM ,which is allowable under the new property regime.

15

Proposition 5 ∂2V∂wM∂wF

is equal to:

1. Before the passage of a property law:

∂2V

∂wM∂wF=

−2

(wM + wF )2+

∂2ψ

∂wM∂wF

2. After the passage of property law, wM/wF ≤ Ω0:

∂2V

∂wM∂wF=

−λ(wM + λwF )2

+∂2ψ

∂wM∂wF

3. After the passage of a property law, wM/wF > Ω0:

∂2V

∂wM∂wF=

−2

(wM + wF )2 +∂2ψ

∂wM∂wF

Proof. See appendix.

This may be positive overall if wM and wF are complementary inputs into the ψ function, which

captures marital utility that does not derive from consumption (but is correlated with premarital

assets). This may happen if people with similar wealth levels are more compatible, due to (for

example) similar tastes or culture. Thus, although husband’s and wife’s wealth are substitutes in

consumption, they may be complementary overall and so we may expect to see positive assortative

mating on wealth.

Again, consider a marriage market with two types of men (wHM , wLM ) and two types of women

(wHF , wLF ). We are interested in whether ∂2V

∂wM∂wFincreases or decreases after the passage of a

property law at all values of wM and wF in this marriage market; this will tell us whether assortative

mating in this market has become more or less valuable. Specifically, we want to know under which

circumstances ∂2V AFTER

∂wM∂wF− ∂2V BEFORE

∂wM∂wF> 0 for all wM , wF in this market.

Lemma 6 There exists a Ω1(λ, π) such that assortative mating will weakly increase when wM/wF ≥

Ω1 and it will strictly decrease when wM/wF < Ω1.

Proof. See appendix.

Practically, consider four sub-marriage markets consisting of men of two wealth types – wHM and

wLM – and women of two wealth types – wHF and wLF . If these wealth levels are such that wM/wF

16

is large in all combinations of wM and wF , we should expect the gains from assortative mating to

increase in this sub-marriage market. However, if wM/wF is small in all combinations, we should

expect the gains from assortative mating to decrease in this sub-marriage market. We do not have

a clear prediction about how the change in the gains from assortative mating should vary across

marriage markets in which wM/wF is large in some combinations but small in others.

To understand what drives this heterogeneity, consider the way in which a married women’s

property law affects borrowing.10 The law has two competing effects: by offering downside protec-

tion in the case of default, the law increases a household’s demand for credit; however, by limiting

the assets a creditor can seize in repayment of a loan, the law decreases the supply of credit. Koudijs

and Salisbury (2016) show that the former effect dominates when a smaller portion of total house-

hold assets is protected (or when wM/wF is large), while the latter effect dominates when a larger

portion of total assets is protected (or when wM/wF is small).

The impact of a property law on borrowing is important, because it affects the way in which

consumption in each state responds to wM and wF , which affects the complementarity of wM and

wF . Couples enjoy consumption in both states. Before the passage of a property law, cG and cB

depend on wM and wF in the same way: wM and wF are perfectly substitutable. After the passage

of a law, cB depends only on wF , while cG depends on both wM and wF , which tends to make

wF and wM more complementary. The heterogeneity in the effect of the law on assortative mating

arises from its impact on the substitutability of wM and wF in determining the utility the couple

derives from cG.

Intuitively, ∂2 log cG/∂wM∂wF reflects the way in which the marginal utility (derived from cG)

of wM varies with wF . In general, an increase in wF decreases the responsiveness of log cG to wM

because the marginal utility of consumption is declining. This is more pronounced when cG is

smaller; to see this, notice that as cG → ∞, log cG approaches zero concavity.11 When wM/wF is

large, the law results in an increase in cG, as it results in an increase in borrowing (and changes in

cG work through changes in borrowing), which tends to diminish the substitutability of wM and

wF . On the other hand, when wM/wF is small, the law results in a decrease in cG, which tends

10See Koudijs and Salisbury (2016) for a detailed investigation into the effect of these laws on total householdinvestment.

11Another way to think about this is the following. For those with low levels of wealth, there are large aggregateutility gains from “diversifying” wM and wF , to smooth consumption across couples. For those with high levels ofwealth, these gains are smaller.

17

to increases the substitutability of wM and wF . For small enough values of wM/wF , this increase

in substitutability overtakes the increase in complementarity driven by the post-law segregation of

wM in cG and wF in cB.

4 Data and Measurement

We link data across two sources: county records of marriages contracted in the South between 1840

and 1851 from familysearch.org; and the complete count 1840 census from ancestry.com. We begin

by extracting information from approximately 300,000 marriage records from southern states dated

between 1840 and 1851 from the genealogical website familysearch.org. These electronic records

contain the full name of both the bride and the groom, the date of marriage, and the county of

marriage. We are able to find marriage records from nine states: Alabama, Arkansas, Georgia,

Kentucky, Louisiana, Mississippi, North Carolina, Tennessee, and Virginia. Table 2 contains in-

formation about the number of marriage records from each state, as well as the coverage of these

records. As can be seen from columns (2) and (3) of this table, our marriage record data cover a

majority of counties in most states.

The second data source is the complete count 1840 census. We use this to measure the pre- mar-

riage socioeconomic status of husbands and wives. The only socioeconomic information available

in the 1840 census is slaveholdings. Specifically, each 1840 census record is taken at the household

level, and contains information on the name of the household head as well as the number of free

and enslaved persons residing in the household. So, we calculate log 1840 household slave wealth

as:

w1840 = log 377S + 1

Here, S is the number of slaves in the household. We multiply this by the average slave price in

1840, which is $377, measured in current dollars (Historical Statistics 2006).

Because we do not have detailed demographic (or even first name) information on household

members, it is difficult to link our couples to their precise 1840 households. Instead, we compute a

measure of “familial assets” by averaging log slave wealth by state and surname, and we link this

to our matched sample by state of marriage and surname (using the maiden name from marriage

records for women). So, the pre-marital wealth of person i with surname j who was married in

18

state s will be:

wi,j,s =1

Kj,s

Kj,s∑k=1

wk,j,s

Here, Kj,s is the number of households in state s headed by someone with the surname j. We

match the spelling of surnames exactly, and individuals whose precise surnames can be found in

the 1840 census comprise our core sample. However, if no exact match can be found, we search

the 1840 census for surnames that approximately mach surnames in our marriage records. We

define an approximate match as one that scores above a certain threshold in a test of alphabetic

string similarity.12 Overall, we are able to obtain an estimate of pre-marital wealth for 80% of our

marriage records, 90% of which are matched exactly.

Table 3 contains further information about our measure of premarital wealth. In panel A, we

present summary statistics about log (household-level) slave wealth from the 1840 census (logw1840)

for all slave states. Overall, approximately one third of all households enumerated in the South

owned at least one slave in 1840. The average log value of slave wealth fluctuated across states,

but the distribution looks broadly similar, particularly among states which ultimately make it into

our sample.

In panel B of table 3, we summarize the measure of log slave wealth that we have matched to our

marriage records (logwi,j,s). We do not have marriage record data from Delaware, Maryland, or

South Carolina, so these states are omitted from this panel. Average log slave wealth looks similar in

both panels. However, because wealth in panel B is computed as an average by surname and state,

the distribution of this measure is more compressed. Standard deviations are smaller, maximum

values are lower, and the fraction of individuals with slave wealth equal to zero is smaller. This can

also be seen in figure 1, which plots the distribution of log slave wealth by surname frequency. Panel

A plots the distribution of log slave wealth at the individual level. It is clear that, at the individual

household level, the distribution of log slave wealth does not differ dramatically by name frequency.

Panel B plots the distribution of mean log slave wealth by name frequency; not surprisingly, the

distribution looks very different, with more common names obeying a significantly more compressed

distribution than unique names.13

12We use the Jaro-Winkler algorithm, which is frequently used in the creation of matched samples (Ruggles et al2010).

13We plot the distribution of groom’s wealth in panel B of this figure; however, the distribution of bride’s wealthlooks identical.

19

Given the difference in the distribution of actual wealth and our measure of wealth, it is worth

mentioning some of the properties of our measure of wealth. We are working under the assumption

of zero linkage error. So, if we observe person i with surname j from state s, we assume that this

person’s family is one of the Kj,s households used to compute wi,j,s.14 However, we do allow for

error in the measurement of “true” log wealth (w∗), so that measured wealth (w) is given by:

w = w∗ + ε

First, notice that our wealth measure is “unbiased” in the sense that it does not differ systematically

from w∗i :

E[w∗i − wi,j,k] = E[w∗|J = j, S = s]− E[w∗|J = j, S = s] = 0

We also derive the expected squared deviation of w∗i from wi,j,k, which captures the variance of

our wealth measure, and is a function of Kj,s and other unknown parameters. Suppose that ε is

IID with mean 0 and variance σ2ε , and that E[w∗2|J = j, S = s] = σ2

j,s. Further, suppose that

E[w∗iw∗k|J = j, S = s] = ρj,s for any i , k. Then, it can be shown that:

E[w∗i − wi,j,k]2 =σ2ε

Kj,s+Kj,s − 1

Kj,s(σ2j,s − ρj,s)

Intuitively, this is increasing in the variance of the measurement error term and increasing in

the dispersion of w∗ within surname-state groups. Given that these are unknown parameters,

it is difficult for us to address this empirically. However, notice also that the overall variance

of measurement error also depends on Kj,s. In particular, as Kj,s increases, measurement error

generated by ε becomes less important, but measurement error generated by dispersion within

surname-state groups becomes more important. This is because, as Kj,s increases, wi,j,s starts to

converge to the median w (as can be see in figure 1). This tends to cause the expected squared

deviation of w from wi,j,s to start to grow. In our individual level analysis, we include binned fixed

effects for name frequency, to deal with the possibility that there are level differences in wj,s which

are correlated with Ki,s. We also run specifications in which we overweight uncommon names.

14We are working on incorporating the possibility of linkage error into our characterization of this wealth measure.

20

5 Empirical Approach

5.1 Empirical Approach: Theory

To analyze the impacts of property laws on the marriage market, we follow Choo and Siow (2006),

who propose a simple statistic for measuring the systematic gains from a marriage between a man

and woman of two types. They define µij to be the number of marriages between men of type i

and women of type j; µi0 to be the number of unmarried men of type i; and µ0j to be the number

of unmarried women of type j. In addition, they define αij to be the systematic gross return to

a type i man from marrying a type j woman, relative to being unmarried; similarly, γij is defined

as the systematic gross return to a type j woman from marrying a type i man, relative to being

unmarried. They show that, under conventional distributional assumptions about idiosyncratic

returns to marriage, the following holds:

ln

(µij√µi0µ0j

)=αij + γij

2

So, the systematic gains to a marriage between a man of type i and a woman of type j can be

measured with information on the number of matches between these two types and the number of

individuals of these types who remain unmarried.

We use a variant of this statistic, which is motivated by the nature of our data.15 In particular,

we cannot observe µi0 or µ0j ; we can only observe marriages that actually occur in a particular

state and year. This does not allow us to measure the value of type i, j marriages. However, it

does allow us to compute the relative value of marriages of different types, which allows us to

characterize the effect of married women’s property laws on assortative matching.

We will consider “types” to be defined by premarital wealth, and we will index men and women

in descending order of wealth. So, i < j means that wi > wj . Consider four “types:” i, k, j,

and l. There will be four sub-marriage markets consisting of men of types i and k and women of

types j and l. Suppose that i < k and j < l. We can say there is a tendency towards assortative

matching in this sub-marriage market if matches between (i, j) and (k, l) systematically yield more

value than matches between (i, l) and (k, j). Or, if marriages among like types systematically more

15This variant of the Choo-Siow statistic is also used in Siow (2015).

21

valuable than marriages among unlike types. With this in mind, we define the following:

Ωijkl =µijµklµilµkj

Then, ωijkl − ln Ωijkl will be equal to:

ωijkl =1

2

((αij + αkl − αil − αkj) + (γij + γkl − γil − γkj)

)

Notice that all µi0 and µ0j terms are differenced out, so we can compute this statistic with the data

we have available. The policy change will tend to increase assortative matching in this sub-marriage

market when it increases ωijkl. This is likely to happen if the policy makes spousal wealth more

“complementary.”

5.2 Empirical Approach: Details

We estimate the average impact of married women’s property acts on ω, as defined above. To

accomplish this, we split our marriage records into state-year “marriage markets.” So, all marriages

occurring in, say, Alabama in 1840 are from the same marriage market. We then divide each

marriage market into “bins” based on the husband’s and wife’s premarital wealth: we assign men

and women to one of B wealth quantiles, which differ by state to reflect to fact that different states

have different wealth distributions.16 So, for each marriage market, we define a B × B matrix,

where men’s wealth quantiles are rows and women’s wealth quantiles are columns. Entry (i, j) is

the number of marriages between men in wealth quantile i and women in wealth quantile j.

Each observation is a group of four sub-marriage markets, consisting of men of two types (i

and k) and women of two types (j and l). An example of one observation is illustrated in figure

2 (B = 10, i = 3, k = 7, j = 3, l = 7). In each marriage market, there are∑B−1

b=1 b possible

combinations of i and k, and∑B−1

b=1 b possible combinations of j and l, which means that there are(∑B−1b=1 b

)2observations per marriage market. With 12 years and 9 states, we have 108 marriage

markets in total, which means that we have 108×(∑B−1

b=1 b)2

observations in total. For example,

16In our preferred specification, we rank couples by the arithmetic mean of 1840 slave wealth by surname andstate. More formally, if W = 377 × S, then we rank individuals with surname j from state s according to Wi,j,s =

1Kj,s

∑Kj,s

k=1 Wk,j,s. We estimate our core specification using even numbers of bins, ranging from B = 4 to B = 20.

22

if B = 10, we would have 2,025 sub-marriage markets in each larger marriage market, which would

give us 108× 2, 025 = 218, 700 observations in total.

We estimate the following:

ωijkl,s,t = α+ βLAWs,t + δt+ χs+ φi + φj + φk + φl + uijkl,s,t (1)

This estimates an average effect of a property law on the systematic value of assortative versus

non-assortative matches in all sub-marriage markets. Notice that, because each observation is a

combination of four sub-marriage markets, very small and very large marriage markets will receive

equal weight. This is not ideal: to deal with this, we weight regressions by the total number of

marriages associated with each observation. We cluster by four variables: state-year-bin i, state-

year-bin j, state-year-bin k, state-year-bin l.

This approach is especially useful for uncovering heterogeneity in the effect of married women’s

property laws in different regions of the marriage market. This is because we can estimate equation

(1) on subsets of the entire market. Recall that, if property laws affect the gains from assortative

mating through their impact on a couple’s access to credit, we should expect to see more assortative

mating in marriage markets with wealthier men and poorer women; however, we should expect to

see less assortative mating in marriage markets with poorer men and wealthier women. We can

test for this type of heterogeneity by estimating equation (1) in exactly these types of marriage

markets. In particular, we divide the marriage market into quadrants, and we estimate equation

(1) four times, using only observations from a particular quadrant at a time.

6 Results

6.1 Individual Level Results

Before estimating the model described in section (5), we explore whether, on average, husbands’

and wives’ wealth became more or less correlated after the passage of a married women’s property

act. In figures 3 and 4, we illustrate this graphically. In figure 3, we calculate the correlation

between husband’s and wife’s log wealth in each state and year, and we plot these correlation

coefficients, marking state-years in which a property law had been passed differently from state-

23

years in which a property law had not been passed. While this is not true in all cases, in many

states that passed a property law, husbands’ and wives’ premarital wealth became somewhat more

positively correlated after the passage of a law. Importantly, there does not seem to be much of an

underlying trend in the correlation between husbands’ and wives’ wealth in any state; so, we are

not concerned that apparent effects of property laws on the marriage market are simply picking up

a different pre-existing trend in assortative mating in states that passed property laws. In figure 4,

we illustrate the change in the correlation between husband’s and wife’s wealth in each state after

the passage of a property law.17 In every state except Alabama, there is at least a small increase

the correlation between husbands’ and wives’ wealth.

To test whether this is significant or not, we use two simple approaches. First, we estimate the

following at the state-year level:

ρs,t = α+ βLAWs,t + δt + χs + us,t

Here, ρs,t is the estimated correlation between husband’s and wife’s log wealth in state s in year

t, LAWs,t is an indicator equal to one if state s had passed a property law by year t, δt is a year

fixed effect, and χs is a state fixed effect. Second, we estimate the following:

wMi,s,t = α+ β1LAWs,t + β2wFi,s,t + γ(LAWs,t × wFi,s,t) + δt + χs+

+τtwFi,s,t + ψswFi,s,t + ui,s,t

Here, wMi,s,t is the log wealth of the husband in couple i, married in state s in year t: wFi,s,t is the

log wealth of the wife in this couple. Other variables are defined similarly. The terms τtwFi,s,t and

ψswFi,s,t are interactions between wife’s log wealth and year fixed effects and state fixed effects,

respectively. These interactions allow the correlation between husband’s and wife’s log wealth to

differ by state and year. The main coefficient of interest here is γ: if γ > 0, this means that

property laws are associated with an increase in the assortative mating.

These results are presented in table 4, with state-year level results in panel A and individual-

level results in panel B. Individual-level results are clustered at the state-year level. We estimate

17These are done using the binscatter command in Stata.

24

several alternative specifications. In column (1), we estimate our baseline specification. In column

(2), we add a state-specific linear time trend (plus an interaction with bride’s log wealth in panel

B). In column (3), we include non-exact matches to the 1840 census. In column (4), we add name

frequency fixed effects (or calculate correlations between residual log husband’s and wife’s wealth,

net of name frequency fixed effects). In column (5), we weight our regressions (or calculate a

weighted correlation between husband’s and wife’s wealth) to address the fact that error in the

measurement of a person’s pre-marital wealth is correlated with the commonness of that person’s

surname. Specifically, we compute the following weight for men from state s with surname j and

women from state t with surname k:

λjs,kt =

(1

Kj,s+(Kj,s − 1

Kj,s

)σ2j,s

)−1/2(1

Kk,t+(Kk,t − 1

Kk,t

)σ2k,t

)−1/2

Here, Kj,s is the number of households in state s with surname j, and σ2j,s is the sample variance of

w among households in state s with surname j. This is an attempt at weighting by the inverse of

the geometric mean of the variance of measurement error associated with the husband’s and wife’s

wealth.18

In all cases, we find that the passage of a property law is associated with a significant increase

in assortative mating. Our weighted results suggest that the error inherent in our measure of

premarital wealth attenuates our estimates slightly, but this is not terribly severe.

6.2 Main Results

In figure 5, we plot our estimates of β from equation (1), with 95% confidence bands, dividing

men and women into different numbers of bins. We estimate equation (1) using a number of bins

ranging from 4 to 20. In the left panel, we use our baseline specification (in which we only use

people with exact surname links to the 1840 census); in the right panel, we include fuzzy links

to the 1840 census. Recall that β captures the impact of LAWs,t on ω, or the relative gains to

assortative mating in each sub-marriage market. This is everywhere positive and significant.

The overall increase in assortative matching masks large differences in the effect of the law in

18We do not know σ2ε and have arbitrarily set this equal to 1. We tried an alternative scheme, in which we simply

weighted by(

1Kj,s

)−1/2(1

Kk,t

)−1/2

and got similar results.

25

different portions of the wealth distribution. We are able to capture this with the model described

in section (5). We estimate our model for the entire distribution of bride’s and groom’s wealth; we

also do this for sub-sections of these distributions, which are illustrated graphically in figure 6. We

find significant changes in particular segments of the distribution, which are illustrated in figure

7. There seems to have been an increase in assortative matching among pairs in which the man

is richer than the woman; conversely, there seems to have been a decline in assortative matching

among pairs in which the woman is richer than the man. Among couples in which both men and

women have above (or below) average wealth, there is no significant change in assortative mating.

This is true in the baseline, and in the version in which we include fuzzy links to the 1840 census.

In figures 8 and 9, we estimate specifications from figures 5 and 7 using alternative definitions

of “marriage market.” Instead of defining a marriage market as a state-year, we define it as a

partial state-year. Specifically, we partition states into quarters, and into halves along an east-west

boundary and along a north-south boundary. The results are robust to redefining marriage markets

in this way. All of this suggests that the credit market effects of married women’s property laws

did affect the marriage market.

7 Alternative Mechanisms

We argue that the heterogeneous effects on the marriage market that we uncover are most consistent

with a model in which married women’s property laws affect the marriage market by affecting the

way a couple can access credit. Here, we discuss other potential channels which may be important,

and we argue that, while some generate an increase in assortative mating overall, they are unlikely

to yield the particular heterogeneity we find.

7.1 Moral Hazard

Property laws affect the distribution of marital property upon separation. While divorce was

extremely uncommon during this period, separation was more common (Cvrcek 2009). In the

absence of a married women’s property law, women were legally entitled to nothing if they were

abandoned by their husbands, though they could in principle seek an equitable settlement to remedy

26

this injustice.19 Married women’s property laws typically awarded women unfettered access to their

separate property in the event of desertion by their husbands.

We model this effect as reallocating marital property in the event that the marriage dis-

solves.With probability π, the couple experiences a severe negative shock to φ, which causes the

couple to separate with certainty. If this occurs, the husband and wife will individually consume

a share of marital property which is determined by the property regime under which the couple

operates. If either the husband or the wife is left with nothing upon separation, he or she can

obtain a “settlement” which is equal to a fraction ω of other half of the couple’s property.

We abstract from any interaction with the credit market here. So, if the couple stays together,

they both consume wM + wF . Before a law is passed, if the couple separates, the husband will

consume (1 − ω)(wM + wF ) and the wife will consume ω(wM + wF ). After a law is passed, if the

couple separates, the husband will consume wM and the wife will consume wF . Thus, total marital

utility before a law is passed is equal to

V (wM , wF , φ) = 2(1−π) log(wM+wF )+π log(

(1−ω)(wM+wF ))

+π log(ω(wM+wF )

)+ψ(wM , wF )+φ

And, after a law is passed, total marital utility is equal to

V (wM , wF , φ) = 2(1− π) log(wM + wF ) + π logwM + π logwF + ψ(wM , wF ) + φ

Proposition 7 After the passage of a property law, assortative mating will always strictly increase.

Proof. See appendix.

The reason is intuitive. Total marital utility depends on the consumption of both husbands and

wives if the couple separates. In the absence of a property law, post-separation consumption of both

halves of the couple depends on wM and wF . With a property law, post-separation consumption

of the wife depends only one wF , while post-separation consumption of the husband depends only

on wM . Thus, wM and wF are inherently less substitutable if a property law is in place. This tends

19When the precise application of the common law led to outcomes deemed “unfair,” these could be “remedied”by a court of equity, or chancery court (Warbasse 1987). Alternatively, married women who had been abandonedmay have sought some form of compensation from their husbands using extralegal means, like enforcement by familyor friends. In any case, we assume there is some mechanism to ensure that married women’ who are abandoned donot have zero consumption; otherwise no one would marry.

27

to encourage assortative mating through this channel.

7.2 Bargaining Power

We have emphasized that the property laws we study in this paper devolved minimal economic

power to married women. As such, their effect on intra-household bargaining should be minimal.

Nonetheless, it is possible that women experienced an increase in bargaining power under these

property laws. How would this affect the marriage market? We consider this question using a

simple model based on Doepke and Tertilt (2009). Proofs are forthcoming in the appendix.

7.2.1 Household production decision

In this model, the decision-making unit is a couple, who has one child. Each half of the couple is

endowed with a level of human capital, Hi, where i ∈ M,F. The couple produces output, which

it consumes, according to the following production function:

c0 = A(tHF )α(HM )1−α

Each person is endowed with 1 unit of time, which the husband is assumed to spend exclusively on

production. The wife splits her time between producing output and educating the child; t represents

time spent on production, and e represents time spent on the child’s education. It must be that

t+ e = 1. Parents transmit human capital to their child according to the following function:

H ′ = eθHβFH

1−βM

Assume the child does not marry and produces output for his or her consumption (c1) alone:

c1 = AH ′ = AeθHβFH

1−βM

Husbands and wives have different preferences over c0 and c1. Specifically:

Ui = (1− γi) log c0 + γi log c1 + νi i ∈ M,F

28

We assume that γF > γM , or women place greater weight on their children’s consumption than

men. The term νi represents all other elements of marital utility – including a direct effect of

wealth – which, for the purposes of this model, are assumed not to change after the passage of

a property law. The couple chooses e to maximize a weighted average of UM and UF , where the

weight depends on bargaining power; this amounts to maximizing the following:

U∗ = (1− γ) log c0 + γ log c1

where γ is a weighted average of γM and γF (Doepke and Tertilt 2009).

7.2.2 Marriage market

Critically, we assume that a husband and wife cannot commit to future investments in children’s

education at the time of marriage. However, knowing their relative bargaining position, they can

anticipate what these investments will look like. So, the couple is able to forecast its total utility

after having a child, and intra-couple utility transfers are made immediately to clear the marriage

market. Thus, as in other transferable utility models, a stable set of marriage matches will be those

that generate the greatest total utility. The couple’s total utility as a function of γ is:

U = UM (γ) + UF (γ)

This is to emphasize that, while the decision about how much to invest in the child’s human capital

is made by weighting the husband’s and wife’s utilities according to relative bargaining power, the

calculation of overall marital utility weights the husband and wife equally.

7.2.3 Effect of laws

We are interested in how a change in bargaining power will affect assortative mating on wealth.

Notice that wealth only directly enters the utility function through ν, which we assume not to

change after the introduction of a property law. However, we assume that allocation of wealth

within the household determines bargaining power in a way that changes after a law is passed.

In particular, we assume (for simplicity) that men hold all the bargaining power before a law is

29

passed, so γ = γM . However, after a law is passed, women acquire a bargaining weight equal to

their share of total family wealth: γ = (1 − ω)γM + ωγF , where ω = wFwM+wF

. This will result in

increased assortative mating if the following holds:

∂2

∂wM∂wF

((2− γM − γF ) log

(c∗0(γ)

)+ (γM + γF ) log

(c∗1(γ)

))> 0

(A more detailed and formal exposition is forthcoming.) In the appendix (also forthcoming), we

show that this is generally the case.

The intuition behind this result is straightforward. If the couple was able to commit ex ante

to a level of investment in its child’s human capital, it would choose the amount that maximizes

total marital utility, then intra-couple utility transfers would be made to satisfy individual parties.

Before the passage of a property law, investment in the child’s human capital is always too low (as

men have all the bargaining power). After the passage of a property law, investment in education

is still too low among couples in which men hold most of the family wealth, and too high among

couples in which women hold most of the family wealth. However, couples in which men and women

have equal wealth (and thus equal bargaining power) invest the efficient amount in their child’s

human capital. As such, these couples experience the largest increase in total marital utility after

the passage of a property law. If couples of similar means experience the largest utility gains, they

are more likely to be formed. This means that assortative mating should become more common

after a property law is enacted.

While this model predicts an increase in assortative mating, it does not predict a decline in

assortative mating among couples with relatively rich wives and relatively poor husbands. The

above logic holds everywhere in the wealth distribution: in any marriage market, the couples who

are most evenly matched (in terms of wealth) will experience the largest utility gains. So, the

model predicts an increase in assortative mating everywhere.

7.3 Changing transfers to daughters

It is possible that married women’s property laws induced parents to transfer more assets to their

married daughters, since they were protected from misappropriation by their daughters’ husbands.

For reason discussed in section 3.1, an increase in transfers to daughters tends to increases as-

30

sortative mating, and a decrease in transfers to daughters tends to reduce assortative mating.20

So, to generate the heterogeneity we observe, parents would have to increase transfers to couples

with rich men and poor women; and, they would have to decrease transfers to couples with rich

women and poor men. This seems unlikely if the credit mechanism is not at play. The potential

for misappropriation by the husband seems largest among rich women married to poor men, and

smallest among rich men married to poor women.

7.4 Endogenous laws

Married women’s property laws were passed during a large recession. It is possible that the states

that passed property laws were the hardest hit by the recession, and that the passage of the laws

coincided with the greatest economic turmoil in these states. If this is the case, then 1840 family

wealth may be a poor measure of family wealth at the time of marriage in these states, particularly

among couples married after the passage of a property law. However, this would tend to lower the

apparent rate of assortative mating, not raise it. In addition, there is no reason for endogenous

laws to generate the heterogeneous effects on the marriage market that we document in this paper.

8 Conclusion

This paper offers evidence that pooling property is an important motive for marriage by analyzing

the impact of married women’s property laws on marriage decisions. We focus on laws passed in the

American South during the 1840s, which re-directed wives’ property toward saving and investment,

and limited husbands’ ability to borrow against their wives’ property. As such, they altered the way

in which married couples could pool property and access the credit market without affecting the

productivity of marriage matches. Using a newly compiled database of linked marriage and census

records, we show that these laws had a heterogeneous effect on the marriage market in different

areas of the wealth distribution. Among couples with relatively rich husbands, assortative mating

became more prevalent, while the opposite occurred among couples with relatively rich wives. We

argue that this patterns can be explained by the fact that property laws affected a couple’s ability

to access credit, and that the ability to access credit as a couple was an important component of

20This is straightforward to show analytically, and a proof is available from the authors on request.

31

the gains from marriage.

References

[1] Abramitzky, Ran, Leah Platt Boustan, and Katherine Eriksson (2012). “Europe’s Tired, Poor,Huddled Masses: Self-Selection and Economic Outcomes in the Age of Mass Migration.” Amer-ican Economic Review. 102(5): 1832-1856.

[2] Adachi, Hiroyuki (2003). “A Search Model of Two-Sided Matching under NontransferableUtility.” Journal of Economic Theory. 113(2): 182-198.

[3] Atack, Jeremay and Fred Bateman (1992). “Matchmaker, Matchmaker, Make Me a Match:A General Personal Computer-Based Matching Program for Historical Research.” HistoricalMethods. 25(2):53-65.

[4] Botticini, Maristella and Aloysius Siow (2003). “Why Dowries?” American Economic Review.93: 1385-98.

[5] Chused, Richard (1983). “Married Women’s Property Law: 1800-1850.” Georgetown Law Jour-nal. 71(2): 1359-1425.

[6] Chen, Natalie, Paola Conconi and Carlo Perroni (2013). “Multi-trait Matching and GenderDifferentials in Intergenerational Mobility.” Economics Letters. 120: 292-296.

[7] Chiappori, Pierre-Andre, Bernard Fortin and Guy Lacroix (2002). “Marriage Market, DivorceLegislation, and Household Labor Supply.” Journal of Political Economy. Vol. 110.

[8] Choo, Eugene and Aloysius Siow (2006). “Who Marries Whom and Why.” Journal of PoliticalEconomy. 114(1): 175-201.

[9] Doepke, Matthias and Michele Tertilt (2009). “Women’s Liberation: What’s in it for Men?”’Quarterly Journal of Economics. 124(4): 1541-1591.

[10] Duflo, Esther (2003). “Grandmothers and Granddaughters: Old Age Pensions and Intrahouse-hold Allocation in South Africa.” World Bank Economic Review. 17(1): 1-25.

[11] Fafchamps, Marcel and Agnes Quisumbing (2005a). “Assets at Marriage in Rural Ethiopia.”Journal of Development Economics, 77(1), pp. 1-25.

[12] Fafchamps, Marcel and Agnes Quisumbing (2005b). “Marriage, Bequest, and AssortativeMatching in Rural Ethiopia.” Economic Development and Cultural Change, 53(2), pp. 347-380.

[13] Fernandez, Raquel (2014). “Women’s Rights and Development.” Journal of Economic Growth.19: 37-80.

[14] Ferrie, Joseph P. (1996). “A New Sample of Americans Linked from the 1850 Public Use MicroSample of the Federal Census of Population to the 1860 Federal Census Manuscript Schedules.”Historical Methods. 29: 141- 156.

32

[15] Gale, David and Lloyed S. Shapely (1962). “College Admissions and the Stability of Marriage.”American Mathematical Monthly. 69(1): 9-15.

[16] Geddes, Rick, Dean Lueck, and Sharon Tennyson (2012). “Human Capital Acquisition and theExpansion of Women’s Economic Rights.” Journal of Law and Economics. 55(4): 839-867.

[17] Geddes, Rick and Dean Lueck (2002). “The Gains from Self-Ownership and the Expansion ofWomen’s Rights.” American Economic Review. 92(4): 1079-1092.

[18] Hamilton, Gillian (1999). “Property Rights and Transaction Costs in Marriage: Evidence fromPrenuptial Contracts.” Journal of Economic History. 59(1): 68-103.

[19] Hirtcsh, Gunter J., Ali Hortacsu, and Dan Ariely (2010). “Matching and Sorting in OnlineDating.” American Economic Review. 100(1): 130-163.

[20] Inwood, Kris and Sarah Van Sligtenhorst (2004). “The Social Consequences of Legal Reform:Women and Property in a Canadian Community.” Continuity and Change. 19(1): 165-197.

[21] Kahn, B. Zorina (1996). “Married Women’s Property Laws and Female Commercial Activity.”Journal of Economic History. 56(2): 356-388.

[22] Kelly, John F. (1882). A Treatise on the Law of Contracts of Married Women. Jersey City:D. Linn & Co.

[23] Koudijs, Peter and Laura Salisbury (2016). “Bankruptcy and Investment: Evidence fromChanges in Marital Property Laws in the U.S. South, 1840-1850.” NBER Working Paper no.21952.

[24] McGrane, Reginald C. (1924). The Panic of 1937: Some Financial Problems of the JacksonianEra. Chicago.

[25] Olivetti, Claudia, Daniele Paserman, and Laura Salisbury (2016) “Three-Generation Mobilityin the United States: The role of Paternal and Maternal Grandparents.” NBER Working Paperno. 22094.

[26] Ruggles, Steven, J. Trent Alexander, Katie Genadek, Ronald Goeken, Matthew B. Schroeder,and Matthew Sobek. (2010) Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database]. Minneapolis: University of Minnesota.

[27] Salmon, Marylynn (1982). “Women and Property in South Carolina: The Evidence fromMarriage Settlements, 1730-1830.” William and Mary Quarterly. 39(4): 655-85.

[28] Warbasse, Elizabeth (1987). The Changing Legal Rights of Married Women. 1800-1861. NewYork: Garland.

[29] Weiss, Yoram (1997). “The Formation and Dissolution of Families: Why Marry? Who MarriesWhom? And What Happens Upon Divorce?” Handbook of Population and Family Economics.1: 81-123.

[30] Wells, John C. (1878). A Treatise on the Separate Property of Married Women. Cincinnati:R. Clarke & Co.

33

Tables and Figures

Figure 1: Distribution of Individual and Grouped Slave Wealth Measure

0.2

.4.6

.8D

ensi

ty

0 5 10 15Log slave wealth

Unique names Names occurring 2−20 XNames occurring > 20 X

Distribution of HH Slave Wealth by Commonness of SurnameIndividual Level Data

®

0.1

.2.3

.4.5

Den

sity

0 5 10 15Log slave wealth

Unique names Names occurring 2−20 XNames occurring > 20 X

Distribution of HH Slave Wealth by Commonness of SurnameGrouped data, groom

®

34

Figure 2: Assortative Matching Data: Illustration

1 2 3 4 5 6 7 8 9 10

1

2

3 (i,j) (i,l)

4 Not4Assortative

Groom's' 5Assortative

Bin6

7 (k,j) (k,l)

8

9

10

Bride's'Bin

35

Figure 3: Correlation between Spouses’ Wealth

Before and After Legal Change

0.0

5.1

.15

.2.2

5C

orr.

Brid

e &

Gro

om W

1840 1845 1850Year

Before After

Alabama

0.0

5.1.

15.2

.25

Cor

r. Br

ide

& G

room

W

1840 1845 1850Year

Before After

Arkansas

0.0

5.1

.15

.2.2

5C

orr.

Brid

e &

Gro

om W

1840 1845 1850Year

Before After

Georgia

0.0

5.1

.15

.2.2

5C

orr.

Brid

e &

Gro

om W

1840 1845 1850Year

Before After

Kentucky

®

0.0

5.1

.15

.2.2

5C

orr.

Brid

e &

Gro

om W

1840 1845 1850Year

Before After

Mississippi

0.0

5.1

.15

.2.2

5C

orr.

Brid

e &

Gro

om W

1840 1845 1850Year

Before After

North Carolina

0.0

5.1

.15

.2.2

5C

orr.

Brid

e &

Gro

om W

1840 1845 1850Year

Before After

Tennessee

0.0

5.1

.15

.2.2

5C

orr.

Brid

e &

Gro

om W

1840 1845 1850Year

Before After

Virginia

®36

Figure 4: Correlation between Spouses’ Wealth

Before and After Legal Change

2.42

.62.

83

3.23

.4G

room

’s W

0 2 4 6 8Bride’s W

Before After

Alabama

1.4

1.6

1.8

2G

room

’s W

0 2 4 6 8Bride’s W

Before

Arkansas

2.8

33.

23.

43.

6G

room

’s W

0 2 4 6 8Bride’s W

Before

Georgia

2.2

2.4

2.6

2.8

Gro

om’s

W

0 2 4 6 8Bride’s W

Before After

Kentucky

®

3.4

3.6

3.8

44.

24.

4G

room

’s W

0 2 4 6 8Bride’s W

Before After

Mississippi

22.

53

3.5

Gro

om’s

W

0 2 4 6 8Bride’s W

Before After

North Carolina

1.6

1.8

22.

22.

4G

room

’s W

0 2 4 6Bride’s W

Before After

Tennessee

2.5

33.

54

Gro

om’s

W

0 2 4 6 8Bride’s W

Before

Virginia

®37

Figure 5: Overall Impact on Assortative Mating

−.05

0.0

5.1

5 10 15 20# Bins

Baseline

−.05

0.0

5.1

5 10 15 20# Bins

Add Fuzzy Matches

Whole Matrix, Different Specifications

®

38

Figure 6: Assortative Matching Results: Area Definitions

1 2 3 4 5 6 7 8 9 10

1

2

3

Groom's' 4Bin

5

6

7

8

9

10

Bride's'Bin

Q1:Rich1man,1rich1woman

Q2:Poor1man,1rich1woman

Q3:Rich1man,1poor1woman

Q4:Poor1man,1poor1woman

39

Figure 7: Impact on Assortative Mating by Quadrant

−.2

−.1

0.1

5 10 15 20# Bins

Rich Men + Rich Women

−.2

−.1

0.1

5 10 15 20# Bins

Rich Men + Poor Women−.

2−.

10

.1

5 10 15 20# Bins

Poor Men + Rich Women

−.2

−.1

0.1

5 10 15 20# Bins

Poor Men + Poor Women

®

(a) Baseline

−.2

−.1

0.1

5 10 15 20# Bins

Rich Men + Rich Women

−.2

−.1

0.1

5 10 15 20# Bins

Rich Men + Poor Women

−.2

−.1

0.1

5 10 15 20# Bins

Poor Men + Rich Women

−.2

−.1

0.1

5 10 15 20# Bins

Poor Men + Poor Women

®

(b) Include Fuzzy Matches

40

Figure 8: Overall Impact on Assortative Mating: Alternative Definitions of Marriage Markets

−.05

0.0

5.1

5 10 15 20# Bins

Baseline

−.05

0.0

5.1

4 6 8 10 12 14# Bins

State Quadrants

−.05

0.0

5.1

4 6 8 10 12 14# Bins

State Halves (West/East Partition)

−.05

0.0

5.1

4 6 8 10 12 14# Bins

State Havles (North/South Partition)

®

41

Figure 9: Impact on Assortative Mating by Quadrant

−.2

−.1

0.1

6 8 10 12 14# Bins

Rich Men + Rich Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Rich Men + Poor Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Poor Men + Rich Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Poor Men + Poor Women

®

(a) State Quadrants

−.2

−.1

0.1

6 8 10 12 14# Bins

Rich Men + Rich Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Rich Men + Poor Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Poor Men + Rich Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Poor Men + Poor Women

®

(b) State Halves (West/East Partition)

−.2

−.1

0.1

6 8 10 12 14# Bins

Rich Men + Rich Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Rich Men + Poor Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Poor Men + Rich Women

−.2

−.1

0.1

6 8 10 12 14# Bins

Poor Men + Poor Women

®

(c) State Halves (North/South Partition)

42

Table 1: Dates of Key Married Women’s Property Legislation in the 1840s

State Date Main Protection Wife’s Assets Ability to SellLaw Change Wife’s Assets

Alabama Mar 1, 1848 All property owned at time of marriage, Wife cannot sellor acquired afterwards

Arkansas –Florida Mar 6, 1845 All property owned at time of marriage, Husband and wife can jointly

or acquired afterwards sell real estateGeorgia –Kentucky Feb 23, 1846 Real estate and slaves owned at time Husband and wife can jointly

of marriage, or acquired afterwards sell real estateLouisiana –Mississippi Feb 28, 1846 Real estate owned at time of marriage Husband and wife can jointly

and all other property required for the sell real estate; wife can sellmaintenance of the plantation (incl. individually if required forslaves) maintenance

North Jan 29, 1849 Husband’s interest in the wife’s real Wife’s real estate cannot beCarolina estate (i.e. profits or rents) not liable sold by husband without her

for his debts written consentTennessee Jan 10, 1850 Husband’s interest in the wife’s real Husband cannot sell his

estate (i.e. profits or rents) not liable interest is his wife’s realfor his debts estate

Texas –Virginia –

Notes: We omit Maryland and South Carolina from this Table as we do not have a sufficient number of marriage

records to include these states in our analysis. Due to their French and Spanish heritage, Louisiana and Texas had

community property systems in place that, by default, allowed men and women to have separate estates. Sources:

Kahn (1996), Geddes and Lueck (2002), Warbasse (1987), Kelly (1882), Wells (1878), Chused (1983) and Salmon

(1982).

43

Table 2: Coverage of Marriage Record Data

State #&Marriage&records %&Counties&with&marriage&record&data

%&Population&living&in&counties&with&

marriage&record&dataAlabama 27,934 0.67 0.75Arkansas 7,186 0.49 0.56Georgia 32,756 0.74 0.78Kentucky 50,507 0.64 0.71Louisiana 5,277 0.19 0.37Mississippi 12,838 0.47 0.65North&Carolina 27,564 0.73 0.76Tennessee 95,371 0.65 0.72Virginia 31,292 0.48 0.54

44

Table 3: Summary Statistics: Log Slave Wealth in 1840

State MeanStandard)deviation

Maximum %)W=0 #)Households

South 2.71 3.75 12.74 0.65 738,527Alabama 3.15 3.95 12.08 0.60 56,079Arkansas 1.64 3.15 11.31 0.78 12,696Delaware 0.60 1.97 9.58 0.91 10,369Georgia 3.15 3.91 12.57 0.60 69,592Kentucky 2.40 3.54 11.16 0.68 100,346Louisiana 3.54 3.99 12.21 0.55 29,930Maryland 1.97 3.33 11.63 0.74 57,831Mississippi 3.98 4.08 12.34 0.50 30,948North)Carolina 2.56 3.69 12.74 0.67 87,491South)Carolina 3.81 4.09 12.59 0.53 46,655Tennessee 1.92 3.33 11.99 0.75 106,554Virginia 2.99 3.84 12.71 0.62 130,036

State MeanStandard)deviation

Maximum %)W=0#)Marriage)

recs%)Marriage)recs)linked)to)1840

South 2.62 1.90 11.74 0.14 264,568 0.79Alabama 3.22 2.00 11.41 0.12 25,216 0.80Arkansas 1.77 2.06 10.22 0.34 5,537 0.58Georgia 3.21 1.87 11.34 0.09 30,649 0.86Kentucky 2.47 1.71 10.31 0.14 45,339 0.81Louisiana 3.35 2.57 10.91 0.22 3,699 0.48Mississippi 4.10 2.24 11.55 0.10 11,328 0.77North)Carolina 2.59 1.82 11.74 0.13 25,941 0.85Tennessee 1.98 1.57 10.95 0.16 87,342 0.84Virginia 3.17 1.97 11.46 0.10 29,517 0.88

Panel&A.&Household&data&from&1840&Census

Panel&B.&Data&from&Marriage&Records

45

Table 4: Effect of Married Women’s Property Laws on Assortative Mating

(1) (2) (3) (4) (5)

Dependent Variable

Post Law 0.028*** 0.040*** 0.018** 0.028*** 0.042**(0.009) (0.012) (0.008) (0.009) (0.020)

Observations 108 108 108 108 108R-squared 0.816 0.830 0.842 0.795 0.765

State & Year FE's Y Y Y Y YState-specific linear time trend N Y N N NInclude fuzzy matches N N Y N NName frequency FEs N N N Y NOverweight uncommon names N N N N Y

Dependent variable

Bride's Log Wealth X Post Law 0.031*** 0.034*** 0.021*** 0.031*** 0.043**(0.007) (0.007) (0.006) (0.007) (0.016)

Observations 210,057 210,057 247,920 210,057 210,057R-squared 0.125 0.125 0.113 0.100 0.120

State & Year FE's Y Y Y Y YState-specific linear time trend N Y N N NInclude fuzzy matches N N Y N NName frequency FEs N N N Y NOverweight uncommon names N N N N Y

Panel A. State-Year-Level Regressions

Correlation between Bride's and Grooms's Log Slave Wealth

Panel B. Individual-Level Regressions

Groom's Log Slave Wealth

46

A Proofs

Proof. of Proposition 1.Becker (1981, p. 130) shows that, given a function f(x, y), ∂2f/∂x∂y > 0 implies that f(x2, y2)+

f(x1, y1) > f(x1, y2) + f(x2, y1), where x2 > x1 and y2 > y1. Similarly, ∂2f/∂x∂y < 0 implies thatf(x2, y2) + f(x1, y1) < f(x1, y2) + f(x2, y1). For reference, the proof is as follows.

Suppose ∂2f/∂x∂y > 0. Consider the following function:

∂Q(x1, x2, y)/∂y ≡ (∂f/∂y)(x2, y)− (∂f/∂y)(x1, y)

If x2 = x1, then ∂Q/∂y = 0. Because ∂Q/∂y increases in x2 by assumption, it follows that∂Q/∂y > 0 when x2 > x1. As the function Q ≡ f(x2, y) − f(x1, y) is increasing in y, it followsthat:

f(x2, y2)− f(x1, y2) > f(x2, y1)− f(x1, y1)

as y2 > y1. Rearranging the above inequality, we get

f(x2, y2) + f(x1, y1) > f(x1, y2) + f(x2, y1)

An identical proof shows that ∂2f/∂x∂y < 0⇒ f(x2, y2) + f(x1, y1) < f(x1, y2) + f(x2, y1).Now, define f(x, y) ≡ U(x, y)− U(x, y), so that f(wHM , w

HF ) ≡ UHH − UHH , and so on. Notice

that∂2f(x, y)

∂x∂y=∂2U(x, y)

∂x∂y− ∂2U(x, y)

∂x∂y

So:∂2f(x, y)

∂x∂y> 0⇔ ∂2U(x, y)

∂x∂y>∂2U(x, y)

∂x∂y

and∂2f(x, y)

∂x∂y< 0⇔ ∂2U(x, y)

∂x∂y<∂2U(x, y)

∂x∂y

Then, by Becker (1981):

∂2U(x, y)

∂x∂y>∂2U(x, y)

∂x∂y⇒ (UHH − UHH) + (ULL − ULL) > (ULH − ULH) + (UHL − UHL)

⇒ UHH + ULL − ULH − ULL > UHH + ULL − ULH − UHL

By a similar proof,

∂2U(x, y)

∂x∂y<∂2U(x, y)

∂x∂y⇒ (UHH − UHH) + (ULL − ULL) < (ULH − ULH) + (UHL − UHL)

⇒ UHH + ULL − ULH − ULL < UHH + ULL − ULH − UHL

Proof. of Proposition 2.Because the household is risk averse, it will always choose a risk-free loan, which is always

incentive compatible. Thus, the husband will choose l∗ that solves the following maximization

47

problem:

maxl

1

2log(R(wM + wF + l)− l

)+

1

2log(R(wM + wF + l)− l

)The first order condition simplifies to:

1

2

(R− 1

R(wM + wF + l)− l− 1−RR(wM + wF + l)− l

)= 0

⇒l∗ =RR− r

(R− 1)(1−R)(wM + wF )

The denominator is positive, from the assumption that R > 1 and R < 1. And, the numerator ispositive so long as RR − r > 0, which we assume. If RR − r < 0, the project is too risky for thehousehold to want to invest, and the household chooses instead to hold all its assets in cash. Weabstract away from this possibility here.

Proof. of Lemma 3.Solutions for ci, where i ∈ G,B, follow straightforwardly from substituting the above solution

for l∗ into the definitions of ci.

Proof. of Proposition 4.Case 1: Risky Loan

We first solve for the case in which the household contracts a risky loan. If we impose thelender’s zero profit condition and take the derivative of UM with respect to l, we get the following:

∂UM∂l

=(2r − 2)

2cG> 0

If the husband’s utility is always increasing in l, he will want to borrow an infinite amount, andthe incentive compatibility constraint will hold with equality. Combining the IC and the lender’szero profit condition, we obtain the following:

R(wM + l)− β∆r(wM + l) = ρl = 2l −R(wM + l)

⇒l =2r − β∆r

2− 2r + β∆rwM

We then obtain the following solution for cG:

cG = R(wM + wF + l)− 2l +R(wM + l) =2β∆r

2− 2r + β∆rwM +RwF

We define the following parameter:

λ ≡ R(2− 2r + β∆r)

2β∆r

48

Notice that λ < 1/2:

λ ≡ R(2− 2r + β∆r)

2β∆r=R(−2(r − 1))

2β∆r+R

2<R(−2(r − 1))

2∆r+R

2

=R(∆r − 2r + 2)

2∆r=R(2− 2R)

2∆r=R−RR

∆r

<R− r

∆r=

1/2(∆r)

∆r=

1

2

This follows from our two restrictions on β and r. Then, we can rewrite cG as follows:

cG =R

λ(wM + λwF )

Because the loan is risky, cB = RwF , or the return on the investment of wF .Case 2: Risk-free Loan

We now solve for the case in which the husband contracts a risk-free loan. In this case, thehusband solves the following maximization problem, subject to the constraint that (1−R)l < RwM :

maxl

1

2log(R(wM + wF + l)− l

)+

1

2log(R(wM + wF + l)− l

)As in the case with no separate property for women, the solution to this (unconstrained) problemis:

l∗ =RR− r

(R− 1)(1−R)(wM + wF )

If the constraint that (1 − R)l < RwM is binding, it will hold with equality and cB will be equalto RwF . In this case, cB is identical with a risky or risk-free loan; thus, the husband will choosethe risky loan in which cG is higher. Note that the (1 − R)l < RwM constraint is binding when(1−R)l∗ = RwM :

(1−R)RR− r

(R− 1)(1−R)(wM + wF ) = RwM

⇒(RR− r)(wF + wM ) = (RR−R)wM

⇒wMwF

=2(RR− r)

∆r

At the same time, we know that, as wF → 0, the husband will certainly prefer a risk-free loan,as cB with a risky loan approaches −∞. So, we know that there exists some cutoff Ω0, where

Ω0 >2(RR−r)

∆r , such that the husband will choose a risky loan if wM/wF ≤ Ω0 and he will choose arisk-free loan if wF /wF > Ω0.

Proof. of Proposition 6.Before the passage of a property law, ∂V/∂wM is:

∂V

∂wM=

2

wM + wF+∂ψ(wM , wF )

∂wM

49

So,∂2V

∂wM∂wF=

−2

(wM + wF )2+

∂2ψ

∂wM∂wF

After the passage of a property law, the effect on assortative mating depends on whether thecouple contracts a risky loan or not. If wM/wF ≤ Ω0, so the couple contracts a risky loan, we getthe following expression for ∂V/∂wM :

∂V

∂wM=

1

wM + λwF+∂ψ(wM , wF )

∂wM

So,∂2V

∂wM∂wF=

−λ(wM + λwF )2

+∂2ψ

∂wM∂wF

If wM/wF < Ω0, cG and cB are identical to the pre-law case, we we obtain the same solutionfor our assortative mating parameter:

∂2V

∂wM∂wF=

−2

(wM + wF )2+

∂2ψ

∂wM∂wF

Proof. of Lemma 6.If wM/wF > Ω0, there is no change in assortative mating, as ∂2V/∂wM∂wF does not change.

If wM/wF ≤ Ω0, the change is ∂2V/∂wM∂wF is equal to:

− λ

(wM + λwF )2+

2

(wM + wF )2

Under which circumstances will this change be positive?

−λ(wM + λwF )2

+2

(wM + wF )2> 0

⇒ λ

(wM + λwF )2<

2

(wM + wF )2> 0

⇒λ(wM + wF )2 < 2(wM + λwF )2

⇒λ1/2(wM + wF ) < 21/2(wM + λwF )

⇒wFλ1/2(

1− (2λ)1/2)< wM

(21/2 − λ1/2

)The right hand side of this inequality is always greater than zero, as 2 > λ. The left hand sideis also positive, as λ < 1/2 (see proof of proposition 4). So, the impact of the law on assortativemating will be heterogeneous. Specifically, the degree of assortative mating will increase if thefollowing holds:

wMwF

>λ1/2

(1− (2λ)1/2

)21/2 − λ1/2

However, the degree of assortative mating will decrease if the inequality is reversed.

Ifλ1/2(

1−(2λ)1/2)

21/2−λ1/2 > Ω0, then Ω1 = Ω0. Otherwise, Ω1 =λ1/2(

1−(2λ)1/2)

21/2−λ1/2 . In either case

assortative mating strictly decreases when wM/wF ≤ Ω1 and weakly increases when wM/wF > Ω1.

50

Proof. of Proposition 7.Given the expression for total marital utility before a law is passed, it is straightforward to

show that:∂2V

∂wM∂wF=

−2

(wM + wF )2+

∂2ψ

∂wM∂wF

After a law is passed:∂2V

∂wM∂wF=−2(1− π)

(wM + wF )2+

∂2ψ

∂wM∂wF

Thus, ∂2V/∂wM∂wF is greater after the law is passed if −2(1− π) > −2, which is clearly true (asπ < 1).

51